GitHub - p-ai-org/p-climate: AI-Powered Climate Reconstruction using Proxy Signals
Project Name: AI for Climate Reconstruction through Proxy Data
Project Manager: Hannah Mandell (PO ’23)
Project Members: Karina Walker, David Chen, Gabriel Yang, Rachel Homan, Nathaniel Getachew, Gabriel Konar-Steenberg, Sahil Rane
Contact:
Earth’s climate has been studied since the earliest age of humankind. Even still, this multi-factored system of geologic, hydrologic, and atmospheric interactions remains one of the most nuanced and important fields of study today – particularly in the wake of a rapidly changing climate system.
An important factor in understanding Earth’s climatic future is understanding Earth’s climatic past. Many fields of study have developed under this one notion. For example, dendroclimatology is the science of determining past climates from the annual properties of tree rings. Similar processes can be completed by examining annual aspects of ice cores, sediment cores, and even the skeletal rings of oceanic organisms. Thanks to these processes, science has gained the famous ‘hockey stick’ diagram that allows us to see the dramatic effect of anthropogenic global warming.
'Hockey Stick' Diagram illustrating anthropogenic impact on century-old climate patterns.
Each of these items are described as ‘proxies’ since the measured aspects of the physical, biological, or isotopic records stored in their annual markers directly relate to measurements of the climate system at the time of creation, thus providing paleo-climatologists with proxies to the temperature, salinity, precipitation, or atmospheric composition of the period without physically having to measure them. Proxies are the key to looking into Earth’s climate past.
Many linear regression models have been scientifically developed between the markers in the proxies and physical measurements of a climate system. However, with the climate system being so notably complex and the samples often having considerate ‘noise’ in their data, it is only reasonable to explore the role of machine learning in improving these models and their subsequent climate reconstructions.
The PAGES (Past Global Change) 2k consortium is one of the most comprehensive sources of information and data related to global temperature over the past couple thousand years. It pulls from a plethora of temperature proxies from many distinct sources. Using the diverse proxy datasets for temperature from PAGES this project aims to construct a temperature-focused climate reconstruction by means of Neural Network computing, a relatively novel approach. Although PAGES datasets contain different organisms and proxies, the aim of the project is to use artificial intelligence to identify an overall signal in the measurements despite the various sources of ‘noise’.